The term “automatic speech recognition” refers to the process of turning a speech signal into a useable message. Automatic voice recognition is a pattern recognition application; hence it needs to extract significant features from its signal. Several acoustic analyses to extract the features are discussed in this work. In speech recognition, the traditional approach to analyzing speech signals involves using spectral analysis to convert the signals into feature vectors. The most commonly used feature set for this purpose is the mel-frequency cepstral coefficient (MFCC), which effectively captures the essential characteristics of speech. In neural networks, a feature map is generated when a filter is applied to the input or the previous layer’s output. This filter moves across the entire layer, one pixel at a time, activating neurons as it goes. The activated outputs are then compiled into the feature map. These feature maps are created by applying filters or feature detectors to either the input image or the output of previous layers. Visualizing these feature maps can provide insights into how the convolutional layers of a model internally represent the input data. Dynamic time warping (DTW) is a well-known technique used to find the best alignment between two sequences that vary over time. It is particularly useful for comparing sequences that may differ in speed or timing, under certain constraints.